Research On Diabetic Retinal Image Classification Based On Deep Learning

With the development of computer vision technology in the medical field,computer-aided diagnosis of medical images has become a hot research field for researchers.Among them,the research of image classification and segmentation algorithm for diabetic retinopathy has become an important research direction of medical image computer aided diagnosis.There are many kinds of diabetic retinopathy,and when patients are in the early stage,the symptoms are not obvious and easy to be ignored,so that the best treatment time will be missed,and even serious cases will have the risk of blindness.In addition,as the number of diabetes patients increases year by year,the problem of excessive consumption and unequal distribution of medical resources becomes increasingly prominent.Therefore,it is of great theoretical and practical significance to study the classification and segmentation algorithm of retinopathy images based on computer vision.This paper focuses on the classification of diabetic retinopathy images and the segmentation of diseased vascular images.(1)The convolution layer and pooling layer of the traditional Alex Net network model were improved,and the pre-processed image data training set and test set were used to train and test the model,and a retinopathy image classification model with better classification accuracy was obtained.Inspired by Google Net network model,the Inception Net module was replaced by Alex Net ordinary convolution to improve the feature extraction capability of convolution network.After convolution at the fifth layer,the spatial pyramid pooling operation was adopted.Compared with the maximum pooling method,the spatial pyramid pooling reduces the requirements on the input image size.So that the input image can be used as input without stretching adjustment.Compared with the traditional Alex Net network model,the classification accuracy of the improved model is improved to 96.1%,and the accuracy is significantly improved.(2)In this paper,an improved U-NET network model is proposed to segment retinal images due to the complex structure of blood vessels.In U-NET network model,Res Net network residual module is used to replace ordinary convolution,which can effectively solve the problem of network degradation and improve the accuracy in the increased depth.The attention mechanism is added to the jump-layer connection of U-NET model.The experimental results show that the improved U-NET network model has better capability of retinal vascular segmentation.